Alternative splicing remodeling and splicing factor regulatory networks define immune-associated molecular heterogeneity in dilated cardiomyopathy

Figures

• 

  FIGURE 1. Schematic overview of the integrated analytical workflow. DCM, dilated cardiomyopathy; DAS, differential alternative splicing; SF, splicing factor; PSI, percent spliced in; DEG, differentially expressed gene; WGCNA, weighted gene co-expression network analysis.

• 

  FIGURE 2. Global landscape of alternative splicing alterations in DCM. (A) UpSet plot showing the overlap of genes harboring different types of significant DAS events, including skipped exon (SE), alternative 5′splice site (A5SS), alternative 3′splice site (A3SS), retained intron (RI), and mutually exclusive exon (MXE). (B) Stacked bar plot showing the number of DAS events across different AS types. Blue and red bars indicate inclusion-down and inclusion-up events in DCM, respectively. (C) Ridge density plots showing the distribution of ΔPSI values across AS types. ΔPSI was calculated as PSI_DCM − PSI_Control. (D) Volcano plot showing DAS events between DCM and control samples. Red dots indicate inclusion-up events in DCM, blue dots indicate inclusion-down events in DCM, and grey dots indicate non-significant or filtered events. Dashed lines indicate the thresholds used for DAS identification. Values with very high -log10(FDR) were capped for visualization if applicable. (E) PCA based on PSI profiles of significant DAS events, showing the separation between DCM and control samples. (F) Heatmap showing row-scaled PSI values of representative DAS events across DCM and control samples.

• 

  FIGURE 3. Functional relevance and representative splicing events of DCM-associated DAS genes. (A) Functional enrichment dot plot of genes harboring significant DAS events. Dot size represents gene count, color indicates -log10(FDR), and dot shape indicates the database source. (B) DAS gene-functional module matrix showing the relationship between representative DAS genes and DCM-relevant functional modules. AS type and ΔPSI direction are annotated on the left. Green blocks indicate functional module involvement. (C) Representative sashimi-like junction-count schematics for selected high-confidence DAS events in TTN, RYR2, LDB3, and FLNA. Control and DCM groups are shown separately, with junction counts, ΔPSI, and FDR values indicated for each event. (D) PSI distribution of representative DAS events between control and DCM samples. ΔPSI was calculated as PSI_DCM–PSI_Control.

• 

  FIGURE 4. Systemic Profiling of splicing factors in Dilated Cardiomyopathy. (A) PCA visualization of three GEO datasets (GSE3585/GSE42955/GSE79962) pre-normalization; (B) Characterizing post-normalization dataset distribution through PCA dimensionality reduction; (C) Circular genome visualization of splicing factor chromosomal localization; (D) Differential splicing factor dysregulation: Heatmap quantification across DCM versus control cohorts; (E) STRING-based PPI network topology among the 24 splicing factors; (F) Correlations among the 24 splicing factors for DCM and normal samples (*p < 0.05, **p < 0.01, ***p < 0.001).

• 

  FIGURE 5. Splicing factors are associated with immune characteristics of DCM. (A) Comparison of ssGSEA-derived infiltration scores for 15 immune cell types between DCM and normal samples (*p < 0.05, **p < 0.01, and ***p < 0.001). (B) Comparison of ssGSEA-derived scores for 13 immune-related functions between DCM and normal samples. (C) Spearman’s rank correlation matrix between the expression of 24 splicing factors and 15 immune-cell infiltration scores within DCM samples. (D) Spearman’s rank correlation matrix between the expression of 24 splicing factors and 13 immune-related functional scores within DCM samples.

• 

  FIGURE 6. Screening characteristic splicing factors by three machine learning algorithms. (A) Search for the optimal value of mtry for SVM-REF model; (B) Least absolute shrinkage and selection operator (LASSO) regression algorithm to identify characteristic markers; (C) Out-of-bag (OOB) error rate of the random forest (RF) model; (D) Search for the optimal value of mtry for RF model; (E) Variable importance plot for the RF model; (F) Venn diagram illustrating the intersection of genes identified by SVM-RF, Lasso, and RF algorithm; (G) Out-of-fold receiver operating characteristic curve of the consensus classifier generated using repeated nested stratified cross-validation. The classifier achieved an AUC of 0.929 (95% CI, 0.862–0.995).

• 

  FIGURE 7. Characteristic SF-AS network in DCM. (A) Workflow for construction of the characteristic SF–AS network. Expression levels of five characteristic SFs were correlated with PSI values of the top 300 DAS events identified by rMATS, followed by correlation filtering and network pruning. (B) Correlation heatmap showing the associations between characteristic SF expression and representative DAS events. Color indicates the Spearman correlation coefficient (R), and asterisks denote significance levels. (C) Characteristic SF-AS network linking five characteristic SFs to key DAS events and downstream functional modules. Solid red and blue edges indicate strong positive and negative correlations, respectively (|R| > 0.5), whereas dashed edges indicate weaker but significant associations (FDR <0.05). Grey lines represent associations between DAS events and functional modules. Different border colors of DAS-event nodes indicate AS event types. (D) Representative scatter plots showing correlations between characteristic SF expression and PSI values of selected DAS events in DCM samples. Spearman correlation coefficients and FDR values are shown in each panel.

• 

  FIGURE 8. Construction of two splicing factors-based subtypes for DCM. (A) Consensus matrix heatmap defining two subtypes (k = 2) and their correlation area; (B) Consensus CDF curves at k = 2–6. (C) The delta area plot; (D) The tracking plot.

• 

  FIGURE 9. Immune characteristics and biological pathways in two DCM clusters. (A) Principal component analysis (PCA) presenting a notable difference in transcriptomes between the two clusters of DCM; (B) Heatmap of splicing genes between the two DCM clusters; (C) Boxplots quantify distinct expression profiles of the 24 splicing factors in two DCM clusters, *p < 0.05, **p < 0.01, ***p < 0.001 vs. clusterA; (D) GSVA for comparing biological pathways function between clusterA and cluster; (E) The infiltration scores of 15 cells infiltrations between two DCM clusters; (F) The infiltration scores of 13 immune-related functions between two DCM clusters.

• 

  FIGURE 10. DEGs and pathway enrichment between two DCM clusters. (A) Heatmap of DEGs between two DCM clusters; (B) KEGG enrichment analysis of DEGs between two DCM clusters; (C) GO enrichment analysis of DEGs between two DCM clusters. BP, biological process, CC, cellular components, MF, molecular functions; (D) Reactome enrichment analysis of DEGs between two DCM clusters.

• 

  FIGURE 11. Construction of WGCNA network and identification of SFs-related hub genes in DCM. (A) Scale-free topology index analysis and mean connectivity of soft threshold power from 1 to 30. The red line displays the scale-free R2 (0.85); (B) The effect of soft threshold power on mean connectivity; (C) Clustering dendrogram of module eigengenes; (D) Heatmap of the correlation between the module eigengenes and DCM clusters; (E) Gene dendrogram clustered through average linkage hierarchical clustering and merged when the correlation of the modules is >0.8; (F) Protein–protein interaction (PPI) network of genes from the black module.

• 

  FIGURE 12. External dataset (GSE5406) validation of characteristic splicing factors in DCM. (A) ROC curve evaluating the diagnostic performance of the combined five-SF signature. (B–F) ROC curves showing the diagnostic performance of individual characteristic splicing factors, including MBNL1 (B), NOVA1 (C), RBM25 (D), RBM5 (E), and SF1 (F). AUC values and 95% confidence intervals are shown in each panel.